Neural network models have recently made striking progress in natural language processing, but they are typically trained on orders of magnitude more language input than children receive. What can these neural networks, which are primarily distributional learners, learn from a naturalistic subset of a single child's experience? We examine this question using a recent longitudinal dataset collected from a single child, consisting of egocentric visual data paired with text transcripts. We train both language-only and vision-and-language neural networks and analyze the (...) linguistic knowledge they acquire. In parallel with findings from Jeffrey Elman's seminal work, the neural networks form emergent clusters of words corresponding to syntactic (nouns, transitive and intransitive verbs) and semantic categories (e.g., animals and clothing), based solely on one child's linguistic input. The networks also acquire sensitivity to acceptability contrasts from linguistic phenomena, such as determiner-noun agreement and argument structure. We find that incorporating visual information produces an incremental gain in predicting words in context, especially for syntactic categories that are comparatively more easily grounded, such as nouns and verbs, but the underlying linguistic representations are not fundamentally altered. Our findings demonstrate which kinds of linguistic knowledge are learnable from a snapshot of a single child's real developmental experience. (shrink)

We examine two connectionist networks—a fractal learning neural network (FLNN) and a Simple Recurrent Network (SRN)—that are trained to process center-embedded symbol sequences. Previous work provides evidence that connectionist networks trained on infinite-state languages tend to form fractal encodings. Most such work focuses on simple counting recursion cases (e.g., anbn), which are not comparable to the complex recursive patterns seen in natural language syntax. Here, we consider exponential state growth cases (including mirror recursion), describe a new training scheme that seems (...) to facilitate learning, and note that the connectionist learning of these cases has a continuous metamorphosis property that looks very different from what is achievable with symbolic encodings. We identify a property—ragged progressive generalization—which helps make this difference clearer. We suggest two conclusions. First, the fractal analysis of these more complex learning cases reveals the possibility of comparing connectionist networks and symbolic models of grammatical structure in a principled way—this helps remove the black box character of connectionist networks and indicates how the theory they support is different from symbolic approaches. Second, the findings indicate the value of future, linked mathematical and empirical work on these models—something that is more possible now than it was 10 years ago. (shrink)

Rumelhart and McClelland's chapter about learning the past tense created a degree of controversy extraordinary even in the adversarial culture of modern science. It also stimulated a vast amount of research that advanced the understanding of the past tense, inflectional morphology in English and other languages, the nature of linguistic representations, relations between language and other phenomena such as reading and object recognition, the properties of artificial neural networks, and other topics. We examine the impact of the Rumelhart and McClelland (...) model with the benefit of 25 years of hindsight. It is not clear who “won” the debate. It is clear, however, that the core ideas that the model instantiated have been assimilated into many areas in the study of language, changing the focus of research from abstract characterizations of linguistic competence to an emphasis on the role of the statistical structure of language in acquisition and processing. (shrink)

Natural languages make prolific use of conventional constituent‐ordering patterns to indicate “who did what to whom,” yet the mechanisms through which these regularities arise are not well understood. A series of recent experiments demonstrates that, when prompted to express meanings through silent gesture, people bypass native language conventions, revealing apparent biases underpinning word order usage, based on the semantic properties of the information to be conveyed. We extend the scope of these studies by focusing, experimentally and computationally, on the interpretation (...) of silent gesture. We show cross‐linguistic experimental evidence that people use variability in constituent order as a cue to obtain different interpretations. To illuminate the computational principles that govern interpretation of non‐conventional communication, we derive a Bayesian model of interpretation via biased inductive inference and estimate these biases from the experimental data. Our analyses suggest people's interpretations balance the ambiguity that is characteristic of emerging language systems, with ordering preferences that are skewed and asymmetric, but defeasible. (shrink)

Natural languages make prolific use of conventional constituent‐ordering patterns to indicate “who did what to whom,” yet the mechanisms through which these regularities arise are not well understood. A series of recent experiments demonstrates that, when prompted to express meanings through silent gesture, people bypass native language conventions, revealing apparent biases underpinning word order usage, based on the semantic properties of the information to be conveyed. We extend the scope of these studies by focusing, experimentally and computationally, on the interpretation (...) of silent gesture. We show cross‐linguistic experimental evidence that people use variability in constituent order as a cue to obtain different interpretations. To illuminate the computational principles that govern interpretation of non‐conventional communication, we derive a Bayesian model of interpretation via biased inductive inference and estimate these biases from the experimental data. Our analyses suggest people's interpretations balance the ambiguity that is characteristic of emerging language systems, with ordering preferences that are skewed and asymmetric, but defeasible. (shrink)

In this article, we consider infants’ acquisition of foundational aspects of language and emotion through the lens of statistical learning. By taking a comparative developmental approach, we highlight ways in which the learning problems presented by input from these two rich communicative domains are both similar and different. Our goal is to encourage other scholars to consider multiple domains of human experience when developing theories in developmental cognitive science.

It has been suggested that external and/or internal limitations paradoxically may lead to superior learning, that is, the concepts of starting small and less is more (Elman, ; Newport, ). In this paper, we explore the type of incremental ordering during training that might help learning, and what mechanism explains this facilitation. We report four artificial grammar learning experiments with human participants. In Experiments 1a and 1b we found a beneficial effect of starting small using two types of simple recursive (...) grammars: right‐branching and center‐embedding, with recursive embedded clauses in fixed positions and fixed length. This effect was replicated in Experiment 2 (N = 100). In Experiment 3 and 4, we used a more complex center‐embedded grammar with recursive loops in variable positions, producing strings of variable length. When participants were presented an incremental ordering of training stimuli, as in natural language, they were better able to generalize their knowledge of simple units to more complex units when the training input “grew” according to structural complexity, compared to when it “grew” according to string length. Overall, the results suggest that starting small confers an advantage for learning complex center‐embedded structures when the input is organized according to structural complexity. (shrink)

Neural plasticity has been invoked as a powerful argument against nativism. However, there is a line of argument, which is well exemplified by Pinker and more recently by Laurence and Margolis The conceptual mind: new directions in the study of concepts, MIT, Cambridge, 2015) with respect to concept nativism, according to which even extreme cases of plasticity show important innate constraints, so that one should rather speak of “constrained plasticity”. According to this view, cortical areas are not really equipotential, they (...) perform instead different kinds of computation, follow essentially different learning rules, or have a fixed internal structure acting as a filter for specific categories of inputs. We intend to analyze this argument, in the light of a review of current neuroscientific literature on plasticity. Our conclusion is that Laurence and Margolis are right in their appeal to innate constraints on connectivity—a thesis that is nowadays welcome to both nativists and non-nativists —but there is little support for their claim of further innate differentiation between and within cortical areas. As we will show, there is instead strong evidence that the cortex is characterized by the indefinite repetition of substantially identical computational units, giving rise in any of its portions to Hebbian, input-dependent plasticity. Although this is entirely compatible with the existence of innate constraints on the brain’s connectivity, the cerebral cortex architecture based on a multiplicity of maps correlating with one another has important computational consequences, a point that has been underestimated by traditional connectionist approaches. (shrink)

Absolute linguistic universals are often justified by cross-linguistic analysis: If all observed languages exhibit a property, the property is taken to be a likely universal, perhaps specified in the cognitive or linguistic systems of language learners and users. In many cases, these patterns are then taken to motivate linguistic theory. Here, we show that cross-linguistic analysis will very rarely be able to statistically justify absolute, inviolable patterns in language. We formalize two statistical methods—frequentist and Bayesian—and show that in both it (...) is possible to find strict linguistic universals, but that the numbers of independent languages necessary to do so is generally unachievable. This suggests that methods other than typological statistics are necessary to establish absolute properties of human language, and thus that many of the purported universals in linguistics have not received sufficient empirical justification. (shrink)

The informativity of a computational model of language acquisition is directly related to how closely it approximates the actual acquisition task, sometimes referred to as the model's cognitive plausibility. We suggest that though every computational model necessarily idealizes the modeled task, an informative language acquisition model can aim to be cognitively plausible in multiple ways. We discuss these cognitive plausibility checkpoints generally and then apply them to a case study in word segmentation, investigating a promising Bayesian segmentation strategy. We incorporate (...) cognitive plausibility by using an age-appropriate unit of perceptual representation, evaluating the model output in terms of its utility, and incorporating cognitive constraints into the inference process. Our more cognitively plausible model shows a beneficial effect of cognitive constraints on segmentation performance. One interpretation of this effect is as a synergy between the naive theories of language structure that infants may have and the cognitive constraints that limit the fidelity of their inference processes, where less accurate inference approximations are better when the underlying assumptions about how words are generated are less accurate. More generally, these results highlight the utility of incorporating cognitive plausibility more fully into computational models of language acquisition. (shrink)

In 2006, Perruchet and Pacton (2006) asked whether implicit learning and statistical learning represent two approaches to the same phenomenon. This article represents an important follow‐up to their seminal review article. As in the previous paper, the focus is on the formation of elementary cognitive units. Both approaches favor different explanations on what these units consist of and how they are formed. Perruchet weighs up the evidence for different explanations and concludes with a helpful agenda for future research.

Cognitive Science, Volume 46, Issue 1, January 2022.

This article explores some of the philosophical implications of the Bayesian modeling paradigm. In particular, it focuses on the ramifications of the fact that Bayesian models pre‐specify an inbuilt hypothesis space. To what extent does this pre‐specification correspond to simply ‘‘building the solution in''? I argue that any learner must have a built‐in hypothesis space in precisely the same sense that Bayesian models have one. This has implications for the nature of learning, Fodor's puzzle of concept acquisition, and the role (...) of modeling in cognitive science. (shrink)

Do you know you are not being massively deceived by an evil demon? That is a familiar skeptical challenge. Less familiar is this question: How do you have a conception of knowledge on which the evil demon constitutes a prima facie challenge? Recently several philosophers have suggested that our responses to skeptical scenarios can be explained in terms of heuristics and biases. We offer an alternative explanation, based in learning theory. We argue that, given the evidence available to the learner, (...) it would be rational for the learner to infer an infallibilist conception of knowledge. (shrink)

People draw subtle distinctions in the normative domain. But it remains unclear exactly what gives rise to such distinctions. On one prominent approach, emotion systems trigger non-utilitarian judgments. The main alternative, inspired by Chomskyan linguistics, suggests that moral distinctions derive from an innate moral grammar. In this article, we draw on Bayesian learning theory to develop a rational learning account. We argue that the ‘size principle’, which is implicated in word learning, can also explain how children would use scant and (...) equivocal evidence to interpret candidate rules as applying more narrowly than utilitarian rules. (shrink)

Inductive generalization, where people go beyond the data provided, is a basic cognitive capability, and it underpins theoretical accounts of learning, categorization, and decision making. To complete the inductive leap needed for generalization, people must make a key ‘‘sampling’’ assumption about how the available data were generated. Previous models have considered two extreme possibilities, known as strong and weak sampling. In strong sampling, data are assumed to have been deliberately generated as positive examples of a concept, whereas in weak sampling, (...) data are assumed to have been generated without any restrictions. We develop a more general account of sampling that allows for an intermediate mixture of these two extremes, and we test its usefulness. In two experiments, we show that most people complete simple one-dimensional generalization tasks in a way that is consistent with their believing in some mixture of strong and weak sampling, but that there are large individual differences in the relative emphasis different people give to each type of sampling. We also show experimentally that the relative emphasis of the mixture is influenced by the structure of the available information. We discuss the psychological meaning of mixing strong and weak sampling, and possible extensions of our modeling approach to richer problems of inductive generalization. (shrink)

Christiansen & Chater ignore the many linguistic universals that cannot be reduced to processing or cognitive constraints, some of which we present. Their claim that grammar is merely acquired language processing skill cannot account for such universals. Their claim that all other universal properties are historically and culturally based is a nonsequitur about language evolution, lacking data.

Identifying patterns in the world requires noticing not only unusual occurrences, but also unusual absences. We examined how people learn from absences, manipulating the extent to which an absence is expected. People can make two types of inferences from the absence of an event: either the event is possible but has not yet occurred, or the event never occurs. A rational analysis using Bayesian inference predicts that inferences from absent data should depend on how much the absence is expected to (...) occur, with less probable absences being more salient. We tested this prediction in two experiments in which we elicited people's judgments about patterns in the data as a function of absence salience. We found that people were able to decide that absences either were mere coincidences or were indicative of a significant pattern in the data in a manner that was consistent with predictions of a simple Bayesian model. (shrink)

An important distinction between phonology and syntax has been overlooked. All phonological patterns belong to the regular region of the Chomsky Hierarchy, but not all syntactic patterns do. We argue that the hypothesis that humans employ distinct learning mechanisms for phonology and syntax currently offers the best explanation for this difference.

It is often claimed that the greatest value of the Bayesian framework in cognitive science consists in its unifying power. Several Bayesian cognitive scientists assume that unification is obviously linked to explanatory power. But this link is not obvious, as unification in science is a heterogeneous notion, which may have little to do with explanation. While a crucial feature of most adequate explanations in cognitive science is that they reveal aspects of the causal mechanism that produces the phenomenon to be (...) explained, the kind of unification afforded by the Bayesian framework to cognitive science does not necessarily reveal aspects of a mechanism. Bayesian unification, nonetheless, can place fruitful constraints on causal–mechanical explanation. 1 Introduction2 What a Great Many Phenomena Bayesian Decision Theory Can Model3 The Case of Information Integration4 How Do Bayesian Models Unify?5 Bayesian Unification: What Constraints Are There on Mechanistic Explanation?5.1 Unification constrains mechanism discovery5.2 Unification constrains the identification of relevant mechanistic factors5.3 Unification constrains confirmation of competitive mechanistic models6 ConclusionAppendix. (shrink)

The acquisition of syntactic categories is a crucial step in the process of acquiring syntax. At this stage, before a full grammar is available, only surface cues are available to the learner. Previous computational models have demonstrated that local contexts are informative for syntactic categorization. However, local contexts are affected by sentence-level structure. In this paper, we add sentence type as an observed feature to a model of syntactic category acquisition, based on experimental evidence showing that pre-syntactic children are able (...) to distinguish sentence type using prosody and other cues. The model, a Bayesian Hidden Markov Model, allows for adding sentence type in a few different ways; we find that sentence type can aid syntactic category acquisition if it is used to characterize the differences in word order between sentence types. In these models, knowledge of sentence type permits similar gains to those found by extending the local context. (shrink)

Generative grammarians typically advocate for a rationalist understanding of language acquisition, according to which the structure of a developed language faculty reflects innate guidance rather than environmental influence. This proposal is developed in developmental linguistics by triggering models of language acquisition. Opposing this tradition, various theorists have advocated for empiricist views of language acquisition, according to which the structure of a developed linguistic competence reflects the linguistic environment in which this competence developed. On this picture, linguistic development is accounted for (...) by general statistical learning mechanisms. In this article I shall precisify the debate, provide a clearer picture of what is at stake, and show why an intermediate picture is needed. (shrink)

This paper defends the view that the Faculty of Language is compositional, i.e., that it computes the meaning of complex expressions from the meanings of their immediate constituents and their structure. I fargue that compositionality and other competing constraints on the way in which the Faculty of Language computes the meanings of complex expressions should be understood as hypotheses about innate constraints of the Faculty of Language. I then argue that, unlike compositionality, most of the currently available non-compositional constraints predict (...) incorrect patterns of early linguistic development. This supports the view that the Faculty of Language is com- positional. More generally, this paper presents a way of framing the compositionality debate (by focusing on its implications for language acquisition) that can lead to its even- tual resolution, so it will hopefully also interest theorists who disagree with its main conclusion. (shrink)

Fodor argued that learning a concept by hypothesis testing would involve an impossible circularity. I show that Fodor's argument implicitly relies on the assumption that actually φ-ing entails an ability to φ. But this assumption is false in cases of φ-ing by luck, and just such luck is involved in testing hypotheses with the kinds of generative random sampling methods that many cognitive scientists take our minds to use. Concepts thus can be learned by hypothesis testing without circularity, and it (...) is plausible that this is how humans in fact acquire at least some of their concepts. (shrink)

The capacity to generate recursive sequences is a marker of rich, algorithmic cognition, and perhaps unique to humans. Yet, the precise processes driving recursive sequence generation remain mysterious. We investigated three potential cognitive mechanisms underlying recursive pattern processing: hierarchical reasoning, ordinal reasoning, and associative chaining. We developed a Bayesian mixture model to quantify the extent to which these three cognitive mechanisms contribute to adult humans’ performance in a sequence generation task. We further tested whether recursive rule discovery depends upon relational (...) information, either perceptual or semantic. We found that the presence of relational information facilitates hierarchical reasoning and drives the generation of recursive sequences across novel depths of center embedding. In the absence of relational information, the use of ordinal reasoning predominates. Our results suggest that hierarchical reasoning is an important cognitive mechanism underlying recursive pattern processing and can be deployed across embedding depths and relational domains. (shrink)

If Bayesian Fundamentalism existed, Jones & Love's (J&L's) arguments would provide a necessary corrective. But it does not. Bayesian cognitive science is deeply concerned with characterizing algorithms and representations, and, ultimately, implementations in neural circuits; it pays close attention to environmental structure and the constraints of behavioral data, when available; and it rigorously compares multiple models, both within and across papers. J&L's recommendation of Bayesian Enlightenment corresponds to past, present, and, we hope, future practice in Bayesian cognitive science.

A central goal of modern generative grammar has been to discover invariant properties of human languages that reflect “the innate schematism of mind that is applied to the data of experience” and that “might reasonably be attributed to the organism itself as its contribution to the task of the acquisition of knowledge” (Chomsky, 1971). Candidates for such invariances include the structure dependence of grammatical rules, and in particular, certain constraints on question formation. Various “poverty of stimulus” (POS) arguments suggest that (...) these invariances reflect an innate human endowment, as opposed to common experience: Such experience warrants selection of the grammars acquired only if humans assume, a priori, that selectable grammars respect substantive constraints. Recently, several researchers have tried to rebut these POS arguments. In response, we illustrate why POS arguments remain an important source of support for appeal to a priori structure-dependent constraints on the grammars that humans naturally acquire. (shrink)

Generalization is a fundamental problem solved by every cognitive system in essentially every domain. Although it is known that how people generalize varies in complex ways depending on the context or domain, it is an open question how people learn the appropriate way to generalize for a new context. To understand this capability, we cast the problem of learning how to generalize as a problem of learning the appropriate hypothesis space for generalization. We propose a normative mathematical framework for learning (...) how to generalize by learning inductive biases for which properties are relevant for generalization in a domain from the statistical structure of features and concepts observed in that domain. More formally, the framework predicts that an ideal learner should learn to generalize by either taking the weighted average of the results of generalizing according to each hypothesis space, with weights given by how well each hypothesis space fits the previously observed concepts, or by using the most likely hypothesis space. We compare the predictions of this framework to human generalization behavior with three experiments in one perceptual (rectangles) and two conceptual (animals and numbers) domains. Across all three studies we find support for the framework's predictions, including individual‐level support for averaging in the third study. (shrink)

A paradox at the heart of language acquisition research is that, to achieve adult-like competence, children must acquire the ability to generalize verbs into non-attested structures, while avoiding utterances that are deemed ungrammatical by native speakers. For example, children must learn that, to denote the reversal of an action, un- can be added to many verbs, but not all (e.g., roll/unroll; close/*unclose). This study compared theoretical accounts of how this is done. Children aged 5–6 (N = 18), 9–10 (N = (...) 18), and adults (N = 18) rated the acceptability of un- prefixed forms of 48 verbs (and, as a control, bare forms). Across verbs, a negative correlation was observed between the acceptability of ungrammatical un- prefixed forms (e.g., *unclose) and the frequency of (a) the bare form and (b) alternative forms (e.g., open), supporting the entrenchment and pre-emption hypotheses, respectively. Independent ratings of the extent to which verbs instantiate the semantic properties characteristic of a hypothesized semantic cryptotype for un- prefixation were a significant positive predictor of acceptability, for all age groups. The relative importance of each factor differed for attested and unattested un- forms and also varied with age. The findings are interpreted in the context of a new hybrid account designed to incorporate the three factors of entrenchment, pre-emption, and verb semantics. (shrink)

The ‘information cycle’ is an evolutionary model of the processes of gene/culture maintenance and change. This paper reports the first naturalistic experimental study designed to collect information data that can illuminate the mechanisms of ‘culture’ production and sharing in information cycles. It is an analysis of conversation among university students in Taiwan. A junior class of 32 students utilized pencil and paper diaries to record conversation topics over a three week period. It was expected that some topics of special interest (...) would be shared widely as they were passed among speakers through repeated information cycles. This type of linear transmission did appear. However, of great interest, the majority of widely shared topics were initiated by larger scales of information cycles, it is argued, specifically popular media, education, social media, ritual, and legal scales, which triggered cascades of conversation sharing. This result indicates that the sharing of cultural information relies on statistical, not linear, transmission, in which topics are shared imprecisely but repeatedly, permitting the construction of shared knowledge. Further, it indicates that culture is shared in a hierarchy of transmission cycles, which differ in space, cycle time, input, and copy fidelity. Further, it indicates that culture must be maintained against depreciation by repeated cycling in information cycles. Cultural transmission is indeed a complex, statistical, multi-scaled process that requires a nested model of information cycles as I have proposed. (shrink)

People can be taught to manipulate symbols according to formal mathematical and logical rules. Cognitive scientists have traditionally viewed this capacity—the capacity for symbolic reasoning—as grounded in the ability to internally represent numbers, logical relationships, and mathematical rules in an abstract, amodal fashion. We present an alternative view, portraying symbolic reasoning as a special kind of embodied reasoning in which arithmetic and logical formulae, externally represented as notations, serve as targets for powerful perceptual and sensorimotor systems. Although symbolic reasoning often (...) conforms to abstract mathematical principles, it is typically implemented by perceptual and sensorimotor engagement with concrete environmental structures. (shrink)